[ https://issues.apache.org/jira/browse/YARN-7094?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16146049#comment-16146049 ]
Junping Du commented on YARN-7094: ---------------------------------- Hi [~rkanter], thanks for the patch here. The content of document looks fine but I think it is better to note that current behavior (decommission immediately when RM HA happens) is a known issue and will improved in later release. Otherwise, user would have different expectation and our improvement later will make behavior inconsistent. Thoughts? > Document that server-side graceful decom is currently not recommended > --------------------------------------------------------------------- > > Key: YARN-7094 > URL: https://issues.apache.org/jira/browse/YARN-7094 > Project: Hadoop YARN > Issue Type: Sub-task > Components: graceful > Affects Versions: 3.0.0-beta1 > Reporter: Robert Kanter > Assignee: Robert Kanter > Priority: Blocker > Attachments: YARN-7094.001.patch > > > Server-side NM graceful decom currently does not work correctly when an RM > failover occurs because we don't persist the info in the state store (see > YARN-5464). Given time constraints for Hadoop 3 beta 1, we've decided to > document this limitation and recommend client-side NM graceful decom in the > meantime if you need this functionality (see [this > comment|https://issues.apache.org/jira/browse/YARN-5464?focusedCommentId=16126119&page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel#comment-16126119]). > Once YARN-5464 is done, we can undo this doc change. -- This message was sent by Atlassian JIRA (v6.4.14#64029) --------------------------------------------------------------------- To unsubscribe, e-mail: yarn-issues-unsubscr...@hadoop.apache.org For additional commands, e-mail: yarn-issues-h...@hadoop.apache.org